Why AI won’t wipe out white-collar jobs

EVER SINCE the arrival of ChatGPT in November 2022, artificial intelligence which can, cheaply and almost instantly, turn a short query in plain English into a workable software app or a data-rich slide deck has excited and terrified in equal measure. Bosses hoping to trim costs love it. Their programmers, PowerPoint jockeys and other desk-bound employees fear it.

Now prominent voices in the world economy are weighing in. In the past week or so Kristalina Georgieva, head of the IMF, has warned that AI is “hitting the labour market like a tsunami”. Jamie Dimon, boss of JPMorgan Chase, has forecast that America’s biggest bank would soon need fewer employees. And Dario Amodei, who runs Anthropic, has predicted that the technology his company is at the forefront of developing could wipe out “half of all entry-level white-collar jobs”.

AI could indeed wreak havoc in the white-collar workforce. But rather than make many such jobs less lucrative—let alone redundant—it is likelier to reshape them. The AI office will look less like a robot and more like a cyborg, combining the best of human and computer capabilities: the Six Million Dollar Man rather than Terminator. To see why, consider what has happened to white-collar work in the past three years, how this compares with earlier technological revolutions and what those patterns imply about what comes next.

For all the alarm, white-collar workers are still doing well. Since late 2022 America has added roughly 3m white-collar jobs—which include management, professional, sales and office roles—while blue-collar employment has remained flat (see chart 1). Some occupations regularly cast as AI’s early victims are on a tear. America has 7% more software developers, 10% more radiologists and 21% more paralegals than three years ago. A slowdown in hiring for some entry-level white-collar work detected lately by academic research appears to predate ChatGPT and, as such, may have more to do with rising interest rates and an increasingly unpredictable global business environment.

Professionals’ pay has held up, too. Since late 2022 real (inflation-adjusted) wages in professional and business services (think salespeople, accountants and the like) have increased by 5%. Office and administrative workers earn 9% more. Controlling for education, age, gender, race and several other characteristics, we calculate that white-collar workers now earn a third more than blue-collar ones (see chart 2). That is nearly triple the premium in the early 1980s and has kept rising over the past three years. AI has not, in other words, so far robbed office workers of their abiding pay advantage.

These findings would not surprise historians of technological change. The early years of the computer age were likewise marked by dire predictions of mass displacement. In 1982 Wassily Leontief, a Nobel-prizewinning economist, cautioned that “the relation between man and machine is being radically transformed”, as computers began to take on “first simple and then increasingly complex mental tasks”. In the event, digital automation proved a boon for office work. Since the early 1980s employment in management, professional, sales and office roles has more than doubled, and their pay has risen by about a third after adjusting for inflation.

One reason white-collar work thrived in earlier digital eras is that computers rarely replaced entire jobs in one go. They automated routine and repetitive tasks—those that could be codified into explicit rules and executed by machines. When a job was all routine and repetition, it could disappear (as happened with typists). But most professional roles are bundles of tasks, only some of which could be automated. The result was not replacement but upgrading: computers raised productivity and let human effort be directed towards higher-value activities like analysis and judgment. Air-traffic controllers illustrate the pattern: software helped process flight data, humans retained authority over high-stakes decisions, wages rose.

More important, by raising productivity and cutting costs, computers also widened the range of activities firms could undertake profitably. E-commerce has generated new work in logistics, supply-chain planning and digital payments. Smartphones created app designers. Social media ushered in digital marketers and influencers. The result was sustained growth in white-collar employment. According to Daron Acemoglu of the Massachusetts Institute of Technology and Pascual Restrepo of Boston University, roughly half of America’s employment growth between 1980 and 2010 came from the creation of entirely new occupations.

AI is cleverer than digital technologies of old. But the same logic of technological change seems likely to apply this time round. For one thing, today’s artificially intelligent systems have what AI scholars call “jagged intelligence”, displaying uneven and inconsistent performance. Being good at 95% of a task is not enough when the remaining 5% involves important edge cases and discretion.

Evidence from Anthropic, drawing on millions of anonymised interactions with its models, bears this out. Only around 4% of occupations use AI across three-quarters or more of their tasks; hardly any roles can be automated in full. As with the computer, AI is reducing the cost of specific cognitive activities—drafting text, writing code, gathering information or running standard analyses—rather than replacing whole roles.

Recent labour-market data support this view. We analysed employment and wage trends across more than 100 large white-collar occupations in America since the second half of 2022. Employment across the sample has risen by 4% and real wages by 3%. To get a sense of AI’s impact on different roles, we used occupational descriptions to classify white-collar roles into four groups depending on the bundles of tasks involved: technical specialists, managers and co-ordinators, care workers, and back-office employees. We then tracked employment in each group starting in late 2022, using six-month moving averages.

Roles that combine technical expertise with oversight and co-ordination have enjoyed the biggest gains. Employment among project managers and information-security experts has risen by 30% or so. Other occupations which combine deep expertise in maths-related fields with problem-solving are also thriving (see chart 3). So are jobs which involve interpersonal care work and those which demand judgment and co-ordination (see chart 4). Only routine back-office work has shrunk. Over the past three years or so the ranks of American insurance-claims clerks have shrunk by 13% and those of secretaries and admin assistants by 20%.

AI is already generating all-new jobs, too. Companies are hiring “data annotators” to label digital information so that AI can parse it, “forward-deployed engineers” to guide clients through AI implementation and, in the C-suite, “chief AI officers”. Indeed, the fastest-growing white-collar occupations in recent years have been those without settled names. “Other mathematical-science occupations” have seen their ranks swell by around 40% since late 2022 and their real wages by about a fifth. “Other computer occupations”, such as systems architects and IT project managers, have also expanded briskly. Employment among “business operations specialists, all other”—a grab-bag combining process design, co-ordination and analysis—has jumped by almost 60%, with solid wage growth to match.

That is not to say all white-collar workers can sleep easy. In the subset of task bundles that include few edge cases and little discretion, AI may soon be able to automate the lot. Already newer models can carry out multi-hour stretches of autonomous work, combining coding, analysis and tool use with limited human input.

Benchmarks created by METR, a research group, suggest that AI can write software by itself for five hours straight, and that this figure has been doubling roughly every seven months. Mr Amodei of Anthropic has mused that AI may be able to do much of a software engineer’s job as soon as this year.

Entry-level jobs look vulnerable for similar reasons. So do those on the receiving end of earlier technological upheavals. The share of Americans in clerical and administrative work, already down from 18% in the 1980s to 10%, looks poised to keep shrinking. New research by Sam Manning and Tomás Aguirre, both of the Centre for the Governance of AI, a think-tank, suggests that such workers have the weakest capacity to adapt, with fewer transferable skills and less scope to move into higher-value jobs.

Such disruption will be painful for the disrupted. But it is a far cry from the sort of labour-market mayhem prophesied by some. Marrying human judgment with machine intelligence is likely to produce more value than AI alone for a while yet. Human accountability and involvement will keep commanding a premium in the marketplace. And white-collar workers have shown themselves to be highly adaptable. AI will redraw their jobs yet again. But it won’t erase them. ■